Thermochimica Acta 474 (2008) 81–86

Contents lists available at ScienceDirect

Thermochimica Acta
journal homepage: www.elsevier.com/locate/tca

A simple and precise linear integral method for isoconversional data

A. Ortega
Instituto de Ciencia de Materiales de Sevilla, Centro Mixto C.S.I.C/Universidad de Sevilla, Avda. Americo Vespucio s/n, 41092 Sevilla, Spain

a r t i c l e i n f o a b s t r a c t

Article history: A simple and precise linear integral method to evaluate the activation energy dependence on the extent
Received 28 March 2008 of conversion has been proposed. The method leads to consistent results with those from differential
Received in revised form 9 May 2008 and integral non-linear procedure (Vyazovkin method). Moreover, the new procedure yields the pre-
Accepted 12 May 2008
exponential factor and the kinetic model. The method was evaluated from isothermal, non-isothermal
Available online 17 May 2008
and non-linear non-isothermal data (CRTA).
© 2008 Elsevier B.V. All rights reserved.
Keywords:
Isoconversional methods
Kinetics parameters
Average Method
Compensation effect
Controlled rate thermal analysis (CRTA)
Isothermal and non-isothermal kinetics

1. Introduction trary heating program. The differential isoconversional method of

The wide applicability of the multiple heating-rate methods
that are based on several experiments has been proved in the
past few years [1–9]. The uncertainty in the calculated value of
the activation energy is significantly reduced by using the most
common representatives of the multiple heating-rate methods, the
isoconversional methods. The merit of such procedures is that the
kinetic law (i.e. the mathematical explicit form of the f(˛) function,
see Table 1) can be ignored completely. Moreover, the isoconver-
sional methods, based on the fact that the reaction rate is only
a function of the temperature for a fixed value of ˛, allow eval-
uating the dependence of the activation energy on the degree
of conversion. Another important success of the isoconversional
methods is that when competitive and independent reaction pro-
ceed concurrently in a system, the mechanism of the reaction may
be revealed by either increasing or decreasing the heating rate
[10].
Budrugeac and Segal [11] have criticized the isoconversional
method and they state that when the activation energy depends
on the degree of conversion, its value obtained by isoconversional
differential and integral methods are different, and that the inte-
gral isoconversional method produces a systematic error in the
activation energy. Vyazovkin [12] has developed an advanced and
rigorous non-linear procedure for data obtained under any arbi-
Friedman [13] presents the most straightforward way to evaluate
the activation energy as a function of the extent of the reaction.
The Friedman method make no mathematical approximations, for
this reason the systematic error in the activation energy for the
linear integral isoconversional method derived assuming a con-
stant activation energy does not appear in the differential method
of Friedman. For this reason one can estimate the systematic error
of an integral isoconversional method by comparing it against the
Friedman method. But this method employs instantaneous rate val-
ues and it is very sensitive to experimental noise, although the
advent of software with smoothing capabilities has greatly reduced
this disadvantage. This situation is effectively avoided by using
the isoconversional method in its integral form. These methods
are based on a simplified approximation of the temperature inte-
gral. Some approximations of the temperature integral leading to
a linear correlation have been developed in order to easily obtain
the activation energy, but there are many inaccuracies associated
with approximations of the temperature integral. In this paper a
simple, but precise, linear integral procedure is proposed without
associated inaccuracies.
2. Theoretical
The reaction rate of a heterogeneous reaction is described by the
following equation:

d˛ E
 
= Af (˛) exp − (1)
E-mail address: aortega@us.es. dt RT

0040-6031/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.tca.2008.05.003
82 A. Ortega / Thermochimica Acta 474 (2008) 81–86

Table 1
Set of reaction models to describe the reaction kinetics in solid-state reactions

Symbol Model Differential f(˛) function Integral g(˛) function

1 − 1/n
JMA (An ) Nucleation and Growth (n = 0.5, 1, 1.5, 2, 2.5, 3, 4) n(1 − ˛)[−ln (1 − ˛)] [−ln (1 − ˛)]1/n
F1 First-order (1 − ˛) −ln (1 − ˛)
Rn Phase-boundary controlled reaction n = 0, 1/2 and 2/3 (1 − ˛)n (1 − (1 − ˛) 1 − n )/1 − n
D1 1D-diffusion 1/2˛ ˛2
D2 2D-diffusion −1/ln (1 − ˛) ˛ + (1 − ˛)ln (1 − ˛)
D3 3D-diffusion (Jander equation) [3(1 − ˛)2/3 ]/[2[1 − (1 − ˛)1/3 ]] [1 − (1 − ˛)1/3 ]2
D4 3D-diffusion (Ginstling-Brounshteinn equation) 3/2[(1 − ˛)−1/3 − 1] 1 − 2˛/3 − (1 − ˛)2/3

the parameters, in this equation, have the usual significance. The This expression also leads to a linear isoconversional procedure,
reaction models, f(˛), most frequently used in solid-state reaction thus, from a series of non-isothermal curves (i = 1, . . ., n), for a given
are included in Table 1. It is well known that in several heteroge- conversion, one obtains:
neous reactions there is a dependence of the effective activation  
ˇi Ea˛
energy on the extent of the reaction [7,14,15]. Thus, when the iso- ln = constant − (9)
conversional method is used, at a constant extent of conversion ˛, T˛,i 2 RT˛,i
from Eq. (1) one arrives at:
The activation energy is obtained from the plots of the l.h.s of Eq.
 d˛  E˛ (9) vs. the reciprocal of the temperature. The FWO and KAS methods
ln = const. − (2)
dt ˛ RT use the integrated form of the rate equation assuming constant acti-
vation energy and consequently an associated error is unavoidable.
where ln (Af(˛)) = const.
The magnitude of this error should be dependent on the magnitude
This equation represents the differential isoconversional proce-
of the variation of E with ˛ and on the value of x = E/RT. The relative
dure, and was developed by Friedman [13]. Integration of Eq. (1)
error |ıE/E|(%) of the activation energy calculated by FWO method
leads for a linear heating rate, ˇ = dT/dt, to:
can be defined by the following expression if we assumed that E
 t  −E  and A are constant:
g(˛) = A exp
RT (t)
dt (3)
ıE
 Ea 
0 (%) = − 1 100 (10)
E E
the integral form of Eq. (3) is represented as:
By differentiating Eq. (7) we obtain:
AE
g(˛) = p(x) (4)
ˇR ∂ ln ˇ Ea˛
= −1.052 (11)
∂(1/T ) R
p(x) is the temperature integral and is given by:
 ∞ On the other hand from Eq. (4) taking logarithm on obtain:
exp(−x)
p(x) = dx (5)  AE 
x
x2 ln g(˛) = ln + ln p(x) (12)
ˇR
where x = E/RT. This function does not have an exact analytical
solution and a large number of approximate equations have been for a series of non-isothermal curves recorded at several heating
proposed in the literature for performing the kinetic analysis of rates, we obtain from Eq. (12) at a constant conversion:
solid-state reactions. The most popular are those of Coats and Red-
ln ˇ = ln p(x) + constant (13)
fern [16], Doyle [17] and Senum and Yang [18], the last of which
even at x = 5 gives only 0.02% deviation from the exact value of the When the p(x) function in Eq. (13) is calculated exactly by
temperature integral and such deviations do not practically affect numerical integration this equation gives the true activation energy
the values of the activation energy. E. Therefore, by differentiating Eq. (13) we obtain:
The most popular isoconversional methods are the methods
∂ ln ˇ ∂ ln p(x)
labeled FWO [19,20] and KAS [21,22]. The Flynn–Wall and Ozawa = (14)
(FWO) integral isoconversional method is based on the Doyle [17] ∂(1/T ) ∂(1/T )
approximation (Eq. (6)) and is one of the method more extensively Since x = E/RT we deduce that:
used as indicated by more than 435 citations received in the last 5
years ∂ ln ˇ E˛ ∂ ln p(x)
= (15)
∂(1/T ) R ∂x
ln(p(x)) = −5.331 − 1.052x (6)
From Eqs. (11) and (15) we deduce that:
From Eqs. (4) and (6) one obtains, from a series of non-
Ea˛ −1 ∂ ln p(x)
isothermal curves, i = 1, . . ., n, for a given conversion: = (16)
E˛ 1.052 ∂x
Ea˛
ln ˇi = constant − 1.052 (7) From Eq. (10) and (16) it is easy to verify that the relative error
RT˛,i
of the FWO method can be represented in the form:
(Ea stands for approximated activation energy).  
Eq. (7) was derived by assuming a constant activation energy. ıE −1 ∂ ln p(x)
(%) = −1 100 (17)
The FWO isoconversional method gives the activation energy from E 1.052 ∂x
a plot of ln ˇ vs. 1/T. The Kissinger–Akahira–Sunose (KAS) method
Eq. (17) allows calculating the relative error in the activation
was derived using the following p(x) approximation
energy for a given conversion ˛, and we deduce from this equation
exp(−x) that the relative error depend on x and, therefore, on the value of
p(x) = (8)
x2 the activation energy.
 A. Ortega / Thermochimica Acta 474 (2008) 81–86 83

For the KAS method the relative error was calculated following on the use of degree of conversion measured at the same tempera-
the same procedure as above, thus by differentiating Eq. (9) we tures on curves recorded at various heating rates. The Popescu’s
obtain: method is not an isoconversional procedure because it uses, on
∂ ln ˇ Ea˛ each curve, a pair of values of conversion at the same tempera-
=− − 2T (18) ture, not the same value of conversion at different temperature
∂(1/T ) R
values. Unlike the Popescu’s method the Average Method is capable
From Eqs. (14) and (18) we deduced that: of revealing the dependence of the activation energy on the degree
of conversion, because most solid-state reactions are not one-step
Ea˛ ∂ ln(p(x)) 2
=− − (19) reaction, analysis by the Average Method (like the advanced iso-
E ∂x x
conversional procedure [12]) may reveal this complexity detected
and from Eqs. (10) and (19) the relative error in the activation energy as a variation of E with ˛. Moreover, this new method is based on
is given by: integration with respect to time that expands the application area
  to arbitrary heating programs.
ıE ∂ ln(p(x) 2
(%) = − − −1 100 (20)
E ∂x x
2.1.1. Estimation of the pre-exponential factor A and the reaction
we can see from Eq. (20) that the relative error also depend on x and, model g(˛)
therefore, on the value of the activation energy. The error decrease A simple method is suggested for the estimation of A, the
with x and increase with the degree of conversion. We can see that method relies on the application of the compensation effect
the FWO and the KAS methods, which have been derived assuming (ln A˛ = mE˛ + n). Eqs. (25) allows ones to arrive at Eq. (26)
a constant activation energy, are very sensitive to the x = E/RT values.
E˛
Thus, for x = 10 the deviation in the activation energy for the FWO − ln t = − ln g(˛) + mE + n − (26)
RT (t)
method is higher than 13% and for x = 20 the deviation decreases to
5%. For this reason Vyazovkin [12] has developed an advanced non- If we assume the original assumption that the reaction model
linear procedure which uses integration over small time segments remains unchanged, g(˛) will be the same and, from the intercept
as follows, as a result the constancy of E is assumed for only a small (I) of the linear relationship (26) we obtain:
segment ˛:
 t˛  −E  ln g(˛) + I˛ = mE˛ + n (where I stand for intercept) (27)
˛
J[E˛ , T (t˛ )] = exp dt (21) Therefore, the compensation parameters m and n may be calcu-
t˛−˛
RT (t)
lated by fitting the Eq. (27) to the experimental data. Values of m
According to this procedure, for a set of n experiments carried and n have been computed using linear regression, with a choice
out at different heating program, Ti (t), the value of E˛ is determined of the kinetic model of Table 1 that yields the best linear correla-
as the value that minimizes the function: tion in Eq. (27). From m and n values we obtain the pre-exponential
n n J[E˛ , T (t˛ )] factors. This procedure, based on the intercepts, can not be used
i
(22) with the others linear isoconversional integral methods, FWO and
i=1 j=i J[E˛ , Tj (t˛ )]
KAS, because of the imprecision of the linear representation and
The associated error is eliminated in this advanced procedure. the relative error in the activation energy (Eqs. (17) and (20)). Now
we check this procedure with different heating programs.
2.1. Average linear integral method
2.1.2. Linear heating program experiments
When E depend on the degree of conversion the simplified inte- For a linear heating program experiments, we can write:
gral methods that use integration from 0 to current ˛ are not
A E˛
appropriate. The average linear procedure proposed here makes ln g(˛ − ˛, ˛) ≈ ln + ln T − (28)
use of integration over small intervals of ˛ and thus avoids its limi- ˇ RT (t)
tations. Like the advanced isoconversional methods it makes use of With ˇ = T/t.
the control over integration by integrating the rate equation over Thus for a given conversion and a set of experiments performed
small ranges. Thus for small segment ˛ the temperature integral under different linear heating rates ˇi (i = 1, . . .., n):
can be approximated using the “average or mean value” theorem:  
 t˛  −E    ˇi E˛
˛ E˛ ln = constant − (29)
J[E˛ , T (t˛ )] = exp dt ≈ t exp − (23) T˛,i RT˛,i
t˛−˛
RT (t˛ ) RT (t˛ )
The activation energy, for each ˛ value, is obtained from the
For small segments (˛ − ˛, ˛) E and A may be assumed con-
plot of the l.h.s of Eq. (29) vs. the reciprocal of the temperature. The
stant, and from Eq. (3) one obtains:
method is very simple and leads to consistent results with those
 t˛  −E 
˛
from differential or non-linear isoconversional technique (Vya-
g(˛ − ˛, ˛) = A˛ exp dt (24) zovkin method). The precision depends on the precise evaluation
t˛−˛
RT (t)
of T˛ − ˛ and T˛ (or t˛ − ˛ and t˛ ).
From Eqs. (23) and (24):
 E˛
 2.1.3. Isothermal experiments
g(˛ − ˛, ˛) ≈ At exp − (25) For a set of isothermal data the dependence of the activation
RT (t)
energy on the degree of conversion can be calculated from Eq. (25)
For a set of n experiments carried out at different heating for T = constant, thus the following relationship is obtained:
programs the activation energy can be determined from the loga-
rithmic form of Eq. (25). The “average or mean value theorem” was E˛
− ln t˛,i ≈ constant − (30)
used before for Popescu [23] to develop an integral method based RT˛,i
84 A. Ortega / Thermochimica Acta 474 (2008) 81–86
2.1.4. Non-linear non-isothermal experiments (CRTA)
The Average Method can be used for other non-linear
non-isothermal processes where the time dependence of the tem-
perature is not simple, or not known, such as CRTA, stepwise
isothermal analysis (SIA) or high-resolution thermogravimetry
(HRTG). In CRTA experiments the reaction temperatures should be
controlled in such a way that both the reaction rate and the par-
tial pressure of the gases produced or consumed in the reaction are
maintained at any constant previously selected value. Therefore,
Eq. (1) can be expressed in the form:
 −E 
C = Af (˛) exp
RT
where C is the constant reaction rate. It is easy to verify that from
Eq. (25) we obtain, for a given conversion and a set of experiments
performed under different constant reaction rates Ci (i = 1, . . .., n):
E˛
ln(t) = constant +
RT˛,i
where t = ˛/C
The dependence of the activation energy on the conversion ˛ is
obtained from the plot of the l.h.s of Eq. (32) vs. the reciprocal of
the temperature.
2.2. Simulations
It is well known that many solid-state reactions show a depen-
dence of the activation energy on the degree of conversion ˛. The
occurrence of such dependence could be interpreted in terms of
a complex reaction mechanism [15]. For example, a variation of
the effective activation energy may be observed for a process that
involves two parallel first-order reactions with two different acti-
vation energies. The overall rate of this process is given as:
d˛ E1
   E2 
= A1 exp − (1 − ˛) + A2 exp − (1 − ˛)
dt RT RT
For four linear heating rates of 1, 2, 4 and 8 ◦ C min−1
where E1 = 167 kJ mol−1 , A1 = 1012 min−1 , E2 = 351 kJ mol−1 and
A2 = 1026 min−1 , the values of the activation energies obtained from
the Friedman method are shown in Fig. 1. We can see that, in this
case, there is a good linear dependence of the effective activation
energy on the degree of conversion ˛:
E = 169.4 + 95.3˛
Fig. 1. E˛ dependence evaluated for the simulated process (parallel first order reac-
tions) by the Friedman method.
In this paper, we use this result to calculate simulated processes
that involves a linear variation of the effective activation energy
with ˛, such as:
E˛ = Eo + b˛ (where b is a constant) (35)
When E depends on the degree of conversion ˛, the appar-
ent activation parameters (E and A) are correlated through the
compensation effect relationship which is the consequence of the
application of the Arrhenius equation. In accordance with this, only
one value of A correspond to each E value [14,24–26]. Thus, in this
paper the following relationship is used:
(31)
ln A˛ = mE˛ + n = m(Eo + b˛) + n (36)
Different heating programs have been simulated: linear
non-isothermal, isothermal and non-isothermal non-linear exper-
iments such as controlled rate thermal analysis (CRTA). All the data
were simulated for first order reactions, F1, i.e., f(˛) = 1− ˛. The fol-
lowing kinetic parameters have been used:
(32)
• For isothermal experiments: Eo = 100 kJ mol−1 ; b = 30 in Eq. (35).
and m = 0.2, n = −1.0 in Eq. (36) and the following isothermal tem-
peratures: 550, 560, 570 and 580 K
• For non-isothermal experiments: Eo = 50 kJ mol−1 ; b = 30 and
m = 0.1, n = −1.0 and the following heating rates: 1, 2, 3 and
4 K min−1
• For non-linear non-isothermal experiments (CRTA):
Eo = 100 kJ mol−1 ; b = 30 and m = 0.2, n = −1.0 and the fol-
lowing constant reaction rate C: 3 × 10−5 ; 6 × 10−5 ; 9 × 10−5 ;
1.2 × 10−4 min−1 ;
Fig. 2 shows the dependence of the activation energy evaluated
by means of different isoconversional methods for isothermal data
(Eq. (30)).
Fig. 3 shows the dependence of the activation energy on the
degree of conversion evaluated for the non-isothermal data (linear
(33) heating rate (Eq. (29))).
Finally Fig. 4 shows the dependence of the activation energy for
the non-linear non-isothermal experiment, CRTA data (Eq. (32)).
2.2.1. Evaluating the pre-exponential factor and the kinetic model
Isoconversional methods identify the dependence of E on ˛ but
do not yield the pre-exponential factor and the reaction model but
with the Average Method here proposed the estimation of the pre-
exponential factor and the kinetic model have been carried out
(34)
Fig. 2. Dependence of E on ˛, for isothermal data, evaluated by means of the Average
Method (), Advanced Vyazovkin () method and the Friedmann () method. The
solid line represents the E simulated value from Eq. (35).
 A. Ortega / Thermochimica Acta 474 (2008) 81–86 85

Fig. 3. Dependence of E on ˛ for non-isothermal data evaluated by different iso-

conversional procedures: the Average Method (), the Advanced Vyazovkin () Fig. 5. Evaluation of the kinetic model from Eq. (27) (values of I˛ + ln g(˛) against
method, the Friedmann () method (the three are superposed) and the KAS and E˛ for all the kinetic models of Table 1).
FWO methods. The solid line represents the E simulated values from Eq. (35).

using the method proposed above (Eq. (27)). This procedure has
been checking using the simulated non-isothermal data with ˇ = 1,
2, 3 and 4 K min−1 (similar results have been obtained from isother-
mal and CRTA data, these results have not been included for the
sake ob brevity). Fig. 5 shows a plot of Eq. (27) for all the reaction
models, and we can see that, of the reaction models of Table 1,the
first-order reaction model, F1, provide the best linearity, this agree
with the kinetic model assumed for the calculation.
Fitting of the reaction model followed by linear regression anal-
ysis resulted in the following compensation parameters: m = 0.101
and n = −0.998 this agree well with the initial assumptions (m = 0.1
and n = −1). As a result, the pre-exponential factor A can be cal-
culated from the relationship (ln A˛ = mE˛ + n), these results are
included in Fig. 6.
Finally, the new procedure has been checked for a set of parallel
first-order consecutive reactions and the following kinetic param-
eters: E1 = 84 kJ mol−1 ; A1 = 6 × 106 min−1 ; E2 = 335 kJ mol−1 ;
A2 = 1.8 × 1029 min−1 and the heating rates: 1, 2, 3 and 4 K min−1 .
The system of differential equations is given by Eq. (33). The
Fig. 6. Dependence of ln A on ˛ for non-isothermal data evaluated by means of
simulated data have been processed by the differential method the Average Method () (ln A˛ = mE˛ + n), solid circles indicate the simulated data
of Friedmann and by integral isoconversional methods. The FWO assumed for the calculation of the theoretical curves.
and KAS methods give rise to a dependence of E on ˛ which

Fig. 4. Dependence of E on ˛, for CRTA data, evaluated by means of the Average

Method (), Advanced Vyazovkin () method and the Friedmann() method (the Fig. 7. Dependence of E on ˛ evaluated by means of isoconversional methods for two
three methods are superposed). The solid line represents the E simulated value from parallel first-order consecutive reactions. The Average Method () the Advanced
Eq. (35). Vyazovkin) method () and the Friedmann () method are superposed.
86 A. Ortega / Thermochimica Acta 474 (2008) 81–86
Fig. 8. Comparison of the simulated non-isothermal TG curve at a heating rate of
ˇ = 2, solid line, with the corresponding TG curve generated using the kinetic triplets
calculated from the Average Method (dotted line).
deviates from the dependence estimated by the others methods.
The Average Method yields a dependence which is practically
identical to that estimated by the differential and the Vyazovkin
methods. These dependences are shown in Fig. 7.
2.2.2. Kinetic predictions
It has been shown that the isoconversional method reveals the
dependence of E on the degree of conversion, this dependence is
a source of kinetic information and helps, not only to disclose the
complexity of the process, but also to identify its kinetic model as
well.
Knowing values of the kinetic triplet one can predict a depen-
dence of ˛ on T. To make kinetic predictions it is necessary to
generate data from the kinetic triplet determined here. The kinetic
parameters obtained from non-isothermal data using the Average
Method have been used to generate data for comparison with initial
simulated data. From the values of the kinetic triplets (E, A and g(˛))
obtained from the Average Method we generate in Fig. 8 a thermo-
analytical curve at a heating-rates of 2 K min−1 . Similar results have
been obtained for the others heating rates.
3. Conclusions
An integral isoconversional procedure have been deduced from
the “average or mean value” theorem to evaluate the activation
energy from isothermal, linear non-isothermal and non-linear non-
isothermal data (CRTA). When the activation energy changes with
the conversion ˛ the Average Method here proposed leads to E
values identical with those obtained by Friedmann and by the
advanced non-linear procedure of Vyazovkin [12]. Moreover, this
procedure, on the contrary that the others linear isoconversional
method, yields the pre-exponential factor and the kinetic model
(with the assumption that the function g(˛) remains the same
when changing temperature). The new procedure is very simple
and precise.
References
[1] M.E. Brown, et al., ICTAC kinetic projet, Thermochim. Acta 355 (2000)
125.
[2] P. Budrugeac, D. Homentcovschi, E. Segal, J. Thermal Anal. Calorim. 63 (2001)
457.
[3] Z. Gao, M. Nakada, I. Amasaki, Thermochim. Acta 369 (2001) 137.
[4] P. Budrugeac, J. Thermal Anal. Calorim. 68 (2002) 131.
[5] T. Wanjun, C. Donghua, Thermochim. Acta 433 (2005) 72.
[6] A. Khawam, D.R. Flanagan, Thermochim. Acta 429 (2005) 93.
[7] A. Khawam, D.R. Flanagan, Thermochim. Acta 436 (2005) 101.
[8] A.K. Burnham, L.N. Dinh, J. Thermal Anal. Calorim. 89 (2007) 2479.
[9] M.J. Starink, Thermochim. Acta 404 (2003) 163.
[10] T. Ozawa, J. Thermal Anal. 9 (1976) 217.
[11] P. Budrugeac, E. Segal, Int. J. Chem. Kinet. 33 (2001) 564.
[12] S. Vyazovkin, J. Comput. Chem. 22 (2001) 178.
[13] J.H. Friedmann, J. Polym. Sci. C 6 (1964-65) 183.
[14] S. Vyazovkin, J.S. Clawson, C. Wight, Chem. Mater. 13 (2001) 960.
[15] S. Vyazovkin, N. Sbirrazzuoli, Macromol. Chem. Phys. 200 (1999) 2294.
[16] A.W. Coats, J.P. Redfern, Nature 201 (1964) 68.
[17] C.D. Doyle, J. Appl. Polym. Sci. 6 (1962) 639.
[18] G.I. Senum, R.T. Yang, J. Thermal Anal. 11 (1979) 445.
[19] T. Ozawa, Bull. Chem. Soc. Jpn. 38 (19651881).
[20] J.H. Flynn, L.A. Wall, Polym. Lett. 4 (1966) 323.
[21] H.E. Kissinger, J Res. Natl. Bur. Std. 57 (1956) 217.
[22] T. Akahira, T. Sunose; Trans. 1969 Joint Convention of Four Electrical Institutes.
Paper No. 246 (19699; Res. Report Chiba Inst. Technol. No. 16,22 (1971).
[23] C. Popescu, Thermochim. Acta 285 (1996) 309.
[24] S. Vyazovkin, Themochim. Acta 128 (1988) 297.
[25] A.K. Galwey, M.E. Brown, Thermal Decomposition of Ionic Solids, Elsevier,
1999.
[26] P. Budrugeac, D.E. Homentcovschi, E. Segal, J. Thermal. Anal. Calorim. 63 (2001)
457.